4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * Copyright (c) 2012, 2013, Intel Corporation.
32 * Use is subject to license terms.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd-zfs/osd_object.c
40 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41 * Author: Mike Pershin <tappro@whamcloud.com>
42 * Author: Johann Lombardi <johann@whamcloud.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_OSD
50 #include <lustre_ver.h>
51 #include <libcfs/libcfs.h>
52 #include <lustre_fsfilt.h>
53 #include <obd_support.h>
54 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_disk.h>
58 #include <lustre_fid.h>
60 #include "osd_internal.h"
62 #include <sys/dnode.h>
67 #include <sys/spa_impl.h>
68 #include <sys/zfs_znode.h>
69 #include <sys/dmu_tx.h>
70 #include <sys/dmu_objset.h>
71 #include <sys/dsl_prop.h>
72 #include <sys/sa_impl.h>
75 static char *osd_obj_tag = "osd_object";
77 static struct dt_object_operations osd_obj_ops;
78 static struct lu_object_operations osd_lu_obj_ops;
79 extern struct dt_body_operations osd_body_ops;
80 static struct dt_object_operations osd_obj_otable_it_ops;
82 extern cfs_mem_cache_t *osd_object_kmem;
85 osd_object_sa_fini(struct osd_object *obj)
88 sa_handle_destroy(obj->oo_sa_hdl);
89 obj->oo_sa_hdl = NULL;
94 osd_object_sa_init(struct osd_object *obj, udmu_objset_t *uos)
98 LASSERT(obj->oo_sa_hdl == NULL);
99 LASSERT(obj->oo_db != NULL);
101 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, obj,
102 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
106 /* Cache the xattr object id, valid for the life of the object */
107 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(uos), &obj->oo_xattr, 8);
109 obj->oo_xattr = ZFS_NO_OBJECT;
112 osd_object_sa_fini(obj);
119 * Add object to list of dirty objects in tx handle.
122 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
124 if (!cfs_list_empty(&obj->oo_sa_linkage))
127 down(&oh->ot_sa_lock);
128 write_lock(&obj->oo_attr_lock);
129 if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
130 cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
131 write_unlock(&obj->oo_attr_lock);
136 * Release spill block dbuf hold for all dirty SAs.
138 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
140 struct osd_object *obj;
142 down(&oh->ot_sa_lock);
143 while (!cfs_list_empty(&oh->ot_sa_list)) {
144 obj = cfs_list_entry(oh->ot_sa_list.next,
145 struct osd_object, oo_sa_linkage);
146 sa_spill_rele(obj->oo_sa_hdl);
147 write_lock(&obj->oo_attr_lock);
148 cfs_list_del_init(&obj->oo_sa_linkage);
149 write_unlock(&obj->oo_attr_lock);
155 * Update the SA and add the object to the dirty list.
157 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
158 void *buf, uint32_t buflen, struct osd_thandle *oh)
162 LASSERT(obj->oo_sa_hdl != NULL);
163 LASSERT(oh->ot_tx != NULL);
165 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
166 osd_object_sa_dirty_add(obj, oh);
172 * Bulk update the SA and add the object to the dirty list.
175 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
176 int count, struct osd_thandle *oh)
180 LASSERT(obj->oo_sa_hdl != NULL);
181 LASSERT(oh->ot_tx != NULL);
183 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
184 osd_object_sa_dirty_add(obj, oh);
190 * Retrieve the attributes of a DMU object
192 int __osd_object_attr_get(const struct lu_env *env, udmu_objset_t *uos,
193 struct osd_object *obj, struct lu_attr *la)
195 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
197 sa_bulk_attr_t *bulk;
202 LASSERT(obj->oo_db != NULL);
204 rc = -sa_handle_get(uos->os, obj->oo_db->db_object, NULL,
205 SA_HDL_PRIVATE, &sa_hdl);
209 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
211 GOTO(out_sa, rc = -ENOMEM);
213 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
214 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
216 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
217 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
219 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
220 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
221 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
222 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
224 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
226 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
230 la->la_atime = osa->atime[0];
231 la->la_mtime = osa->mtime[0];
232 la->la_ctime = osa->ctime[0];
233 la->la_mode = osa->mode;
234 la->la_uid = osa->uid;
235 la->la_gid = osa->gid;
236 la->la_nlink = osa->nlink;
237 la->la_flags = attrs_zfs2fs(osa->flags);
238 la->la_size = osa->size;
240 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
241 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(uos), &osa->rdev, 8);
244 la->la_rdev = osa->rdev;
245 la->la_valid |= LA_RDEV;
248 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
250 sa_handle_destroy(sa_hdl);
255 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
256 uint64_t oid, dmu_buf_t **dbp, void *tag)
258 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
263 rc = -sa_buf_hold(os, oid, tag, dbp);
267 dmu_object_info_from_db(*dbp, doi);
268 if (unlikely (oid != DMU_USERUSED_OBJECT &&
269 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
270 sa_buf_rele(*dbp, tag);
276 LASSERT((*dbp)->db_object == oid);
277 LASSERT((*dbp)->db_offset == -1);
278 LASSERT((*dbp)->db_data != NULL);
284 * Concurrency: no concurrent access is possible that early in object
287 struct lu_object *osd_object_alloc(const struct lu_env *env,
288 const struct lu_object_header *hdr,
291 struct osd_object *mo;
293 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, CFS_ALLOC_IO);
297 l = &mo->oo_dt.do_lu;
298 dt_object_init(&mo->oo_dt, NULL, d);
299 mo->oo_dt.do_ops = &osd_obj_ops;
300 l->lo_ops = &osd_lu_obj_ops;
301 CFS_INIT_LIST_HEAD(&mo->oo_sa_linkage);
302 init_rwsem(&mo->oo_sem);
303 sema_init(&mo->oo_guard, 1);
304 rwlock_init(&mo->oo_attr_lock);
312 * Concurrency: shouldn't matter.
314 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
316 struct osd_device *osd = osd_obj2dev(obj);
317 udmu_objset_t *uos = &osd->od_objset;
318 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
322 if (obj->oo_db == NULL)
327 rc = osd_object_sa_init(obj, uos);
331 /* cache attrs in object */
332 rc = __osd_object_attr_get(env, &osd->od_objset,
337 if (likely(!fid_is_acct(fid)))
338 /* no body operations for accounting objects */
339 obj->oo_dt.do_body_ops = &osd_body_ops;
342 * initialize object before marking it existing
344 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
347 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
352 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
354 struct osd_thread_info *info = osd_oti_get(env);
357 struct lustre_mdt_attrs *lma;
360 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
361 lma = (struct lustre_mdt_attrs *)info->oti_buf;
363 buf.lb_len = sizeof(info->oti_buf);
365 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
368 lustre_lma_swab(lma);
369 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
370 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
371 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
372 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
373 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
374 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
377 } else if (rc == -ENODATA) {
378 /* haven't initialize LMA xattr */
386 * Concurrency: no concurrent access is possible that early in object
389 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
390 const struct lu_object_conf *conf)
392 struct osd_object *obj = osd_obj(l);
393 struct osd_device *osd = osd_obj2dev(obj);
398 LASSERT(osd_invariant(obj));
400 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
401 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
402 l->lo_header->loh_attr |= LOHA_EXISTS;
406 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
408 LASSERT(obj->oo_db == NULL);
409 rc = __osd_obj2dbuf(env, osd->od_objset.os, oid,
410 &obj->oo_db, osd_obj_tag);
412 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
413 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
417 rc = osd_object_init0(env, obj);
421 rc = osd_check_lma(env, obj);
424 } else if (rc == -ENOENT) {
427 LASSERT(osd_invariant(obj));
433 * Concurrency: no concurrent access is possible that late in object
436 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
438 struct osd_object *obj = osd_obj(l);
440 LASSERT(osd_invariant(obj));
442 dt_object_fini(&obj->oo_dt);
443 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
446 static void __osd_declare_object_destroy(const struct lu_env *env,
447 struct osd_object *obj,
448 struct osd_thandle *oh)
450 struct osd_device *osd = osd_obj2dev(obj);
451 udmu_objset_t *uos = &osd->od_objset;
452 dmu_buf_t *db = obj->oo_db;
453 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
454 uint64_t oid = db->db_object, xid;
455 dmu_tx_t *tx = oh->ot_tx;
459 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
461 /* zap holding xattrs */
462 if (obj->oo_xattr != ZFS_NO_OBJECT) {
465 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
467 rc = -udmu_zap_cursor_init(&zc, uos, oid, 0);
471 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
472 BUG_ON(za->za_integer_length != sizeof(uint64_t));
473 BUG_ON(za->za_num_integers != 1);
475 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
476 sizeof(uint64_t), 1, &xid);
478 CERROR("%s: xattr lookup failed: rc = %d\n",
482 dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
484 zap_cursor_advance(zc);
489 udmu_zap_cursor_fini(zc);
492 if (rc && tx->tx_err == 0)
496 static int osd_declare_object_destroy(const struct lu_env *env,
497 struct dt_object *dt,
500 char *buf = osd_oti_get(env)->oti_str;
501 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
502 struct osd_object *obj = osd_dt_obj(dt);
503 struct osd_device *osd = osd_obj2dev(obj);
504 struct osd_thandle *oh;
510 LASSERT(dt_object_exists(dt));
512 oh = container_of0(th, struct osd_thandle, ot_super);
513 LASSERT(oh->ot_tx != NULL);
515 /* declare that we'll destroy the object */
516 __osd_declare_object_destroy(env, obj, oh);
518 /* declare that we'll remove object from fid-dnode mapping */
519 zapid = osd_get_name_n_idx(env, osd, fid, buf);
520 dmu_tx_hold_bonus(oh->ot_tx, zapid);
521 dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
523 /* declare that we'll remove object from inode accounting ZAPs */
524 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
525 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
526 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
527 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
530 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
531 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
535 /* data to be truncated */
536 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
537 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
541 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
543 LASSERT(uos->objects != 0);
544 spin_lock(&uos->lock);
546 spin_unlock(&uos->lock);
548 return -dmu_object_free(uos->os, oid, tx);
552 * Delete a DMU object
554 * The transaction passed to this routine must have
555 * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
556 * and then assigned to a transaction group.
558 * This will release db and set it to NULL to prevent further dbuf releases.
560 static int __osd_object_destroy(const struct lu_env *env,
561 struct osd_object *obj,
562 dmu_tx_t *tx, void *tag)
564 struct osd_device *osd = osd_obj2dev(obj);
565 udmu_objset_t *uos = &osd->od_objset;
567 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
571 /* Assert that the transaction has been assigned to a
572 transaction group. */
573 LASSERT(tx->tx_txg != 0);
575 /* zap holding xattrs */
576 if (obj->oo_xattr != ZFS_NO_OBJECT) {
577 rc = -udmu_zap_cursor_init(&zc, uos, obj->oo_xattr, 0);
580 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
581 BUG_ON(za->za_integer_length != sizeof(uint64_t));
582 BUG_ON(za->za_num_integers != 1);
584 rc = -zap_lookup(uos->os, obj->oo_xattr, za->za_name,
585 sizeof(uint64_t), 1, &xid);
587 CERROR("%s: lookup xattr %s failed: rc = %d\n",
588 osd->od_svname, za->za_name, rc);
591 rc = __osd_object_free(uos, xid, tx);
593 CERROR("%s: fetch xattr %s failed: rc = %d\n",
594 osd->od_svname, za->za_name, rc);
595 zap_cursor_advance(zc);
597 udmu_zap_cursor_fini(zc);
599 rc = __osd_object_free(uos, obj->oo_xattr, tx);
601 CERROR("%s: freeing xattr failed: rc = %d\n",
605 return __osd_object_free(uos, obj->oo_db->db_object, tx);
608 static int osd_object_destroy(const struct lu_env *env,
609 struct dt_object *dt,
612 char *buf = osd_oti_get(env)->oti_str;
613 struct osd_object *obj = osd_dt_obj(dt);
614 struct osd_device *osd = osd_obj2dev(obj);
615 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
616 struct osd_thandle *oh;
621 LASSERT(obj->oo_db != NULL);
622 LASSERT(dt_object_exists(dt));
623 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
625 oh = container_of0(th, struct osd_thandle, ot_super);
627 LASSERT(oh->ot_tx != NULL);
629 zapid = osd_get_name_n_idx(env, osd, fid, buf);
631 /* remove obj ref from index dir (it depends) */
632 rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
634 CERROR("%s: zap_remove() failed: rc = %d\n",
639 /* Remove object from inode accounting. It is not fatal for the destroy
640 * operation if something goes wrong while updating accounting, but we
641 * still log an error message to notify the administrator */
642 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
643 obj->oo_attr.la_uid, -1, oh->ot_tx);
645 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
646 " %d: rc = %d\n", osd->od_svname, PFID(fid),
647 obj->oo_attr.la_uid, rc);
648 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
649 obj->oo_attr.la_gid, -1, oh->ot_tx);
651 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
652 " %d: rc = %d\n", osd->od_svname, PFID(fid),
653 obj->oo_attr.la_gid, rc);
656 rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
658 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
664 /* not needed in the cache anymore */
665 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
670 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
672 struct osd_object *obj = osd_obj(l);
674 if (obj->oo_db != NULL) {
675 osd_object_sa_fini(obj);
676 if (obj->oo_sa_xattr) {
677 nvlist_free(obj->oo_sa_xattr);
678 obj->oo_sa_xattr = NULL;
680 sa_buf_rele(obj->oo_db, osd_obj_tag);
681 cfs_list_del(&obj->oo_sa_linkage);
687 * Concurrency: ->loo_object_release() is called under site spin-lock.
689 static void osd_object_release(const struct lu_env *env,
695 * Concurrency: shouldn't matter.
697 static int osd_object_print(const struct lu_env *env, void *cookie,
698 lu_printer_t p, const struct lu_object *l)
700 struct osd_object *o = osd_obj(l);
702 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
705 static void osd_object_read_lock(const struct lu_env *env,
706 struct dt_object *dt, unsigned role)
708 struct osd_object *obj = osd_dt_obj(dt);
710 LASSERT(osd_invariant(obj));
712 down_read(&obj->oo_sem);
715 static void osd_object_write_lock(const struct lu_env *env,
716 struct dt_object *dt, unsigned role)
718 struct osd_object *obj = osd_dt_obj(dt);
720 LASSERT(osd_invariant(obj));
722 down_write(&obj->oo_sem);
725 static void osd_object_read_unlock(const struct lu_env *env,
726 struct dt_object *dt)
728 struct osd_object *obj = osd_dt_obj(dt);
730 LASSERT(osd_invariant(obj));
731 up_read(&obj->oo_sem);
734 static void osd_object_write_unlock(const struct lu_env *env,
735 struct dt_object *dt)
737 struct osd_object *obj = osd_dt_obj(dt);
739 LASSERT(osd_invariant(obj));
740 up_write(&obj->oo_sem);
743 static int osd_object_write_locked(const struct lu_env *env,
744 struct dt_object *dt)
746 struct osd_object *obj = osd_dt_obj(dt);
749 LASSERT(osd_invariant(obj));
751 if (down_write_trylock(&obj->oo_sem)) {
753 up_write(&obj->oo_sem);
758 static int osd_attr_get(const struct lu_env *env,
759 struct dt_object *dt,
760 struct lu_attr *attr,
761 struct lustre_capa *capa)
763 struct osd_object *obj = osd_dt_obj(dt);
767 LASSERT(dt_object_exists(dt));
768 LASSERT(osd_invariant(obj));
771 read_lock(&obj->oo_attr_lock);
772 *attr = obj->oo_attr;
773 read_unlock(&obj->oo_attr_lock);
775 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
776 * from within sa_object_size() can block on a mutex, so
777 * we can't call sa_object_size() holding rwlock */
778 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
779 /* we do not control size of indices, so always calculate
780 * it from number of blocks reported by DMU */
781 if (S_ISDIR(attr->la_mode))
782 attr->la_size = 512 * blocks;
783 /* Block size may be not set; suggest maximal I/O transfers. */
785 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
787 attr->la_blksize = blksize;
788 attr->la_blocks = blocks;
789 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
794 /* Simple wrapper on top of qsd API which implement quota transfer for osd
795 * setattr needs. As a reminder, only the root user can change ownership of
796 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
797 static inline int qsd_transfer(const struct lu_env *env,
798 struct qsd_instance *qsd,
799 struct lquota_trans *trans, int qtype,
800 __u64 orig_id, __u64 new_id, __u64 bspace,
801 struct lquota_id_info *qi)
805 if (unlikely(qsd == NULL))
808 LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
809 qi->lqi_type = qtype;
811 /* inode accounting */
812 qi->lqi_is_blk = false;
814 /* one more inode for the new owner ... */
815 qi->lqi_id.qid_uid = new_id;
817 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
818 if (rc == -EDQUOT || rc == -EINPROGRESS)
823 /* and one less inode for the current id */
824 qi->lqi_id.qid_uid = orig_id;;
826 /* can't get EDQUOT when reducing usage */
827 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
828 if (rc == -EINPROGRESS)
833 /* block accounting */
834 qi->lqi_is_blk = true;
836 /* more blocks for the new owner ... */
837 qi->lqi_id.qid_uid = new_id;
838 qi->lqi_space = bspace;
839 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
840 if (rc == -EDQUOT || rc == -EINPROGRESS)
845 /* and finally less blocks for the current owner */
846 qi->lqi_id.qid_uid = orig_id;
847 qi->lqi_space = -bspace;
848 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
849 /* can't get EDQUOT when reducing usage */
850 if (rc == -EINPROGRESS)
855 static int osd_declare_attr_set(const struct lu_env *env,
856 struct dt_object *dt,
857 const struct lu_attr *attr,
858 struct thandle *handle)
860 struct osd_thread_info *info = osd_oti_get(env);
861 char *buf = osd_oti_get(env)->oti_str;
862 struct osd_object *obj = osd_dt_obj(dt);
863 struct osd_device *osd = osd_obj2dev(obj);
864 struct osd_thandle *oh;
870 if (!dt_object_exists(dt)) {
871 /* XXX: sanity check that object creation is declared */
875 LASSERT(handle != NULL);
876 LASSERT(osd_invariant(obj));
878 oh = container_of0(handle, struct osd_thandle, ot_super);
880 LASSERT(obj->oo_sa_hdl != NULL);
881 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
883 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
884 bspace = toqb(bspace * blksize);
886 if (attr && attr->la_valid & LA_UID) {
887 /* account for user inode tracking ZAP update */
888 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
889 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
891 /* quota enforcement for user */
892 if (attr->la_uid != obj->oo_attr.la_uid) {
893 rc = qsd_transfer(env, osd->od_quota_slave,
894 &oh->ot_quota_trans, USRQUOTA,
895 obj->oo_attr.la_uid, attr->la_uid,
896 bspace, &info->oti_qi);
901 if (attr && attr->la_valid & LA_GID) {
902 /* account for user inode tracking ZAP update */
903 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
904 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
906 /* quota enforcement for group */
907 if (attr->la_gid != obj->oo_attr.la_gid) {
908 rc = qsd_transfer(env, osd->od_quota_slave,
909 &oh->ot_quota_trans, GRPQUOTA,
910 obj->oo_attr.la_gid, attr->la_gid,
911 bspace, &info->oti_qi);
921 * Set the attributes of an object
923 * The transaction passed to this routine must have
924 * dmu_tx_hold_bonus(tx, oid) called and then assigned
925 * to a transaction group.
927 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
928 const struct lu_attr *la, struct thandle *handle,
929 struct lustre_capa *capa)
931 struct osd_object *obj = osd_dt_obj(dt);
932 struct osd_device *osd = osd_obj2dev(obj);
933 udmu_objset_t *uos = &osd->od_objset;
934 struct osd_thandle *oh;
935 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
936 sa_bulk_attr_t *bulk;
941 LASSERT(handle != NULL);
942 LASSERT(dt_object_exists(dt));
943 LASSERT(osd_invariant(obj));
944 LASSERT(obj->oo_sa_hdl);
946 oh = container_of0(handle, struct osd_thandle, ot_super);
947 /* Assert that the transaction has been assigned to a
948 transaction group. */
949 LASSERT(oh->ot_tx->tx_txg != 0);
951 if (la->la_valid == 0)
954 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
958 /* do both accounting updates outside oo_attr_lock below */
959 if ((la->la_valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
960 /* Update user accounting. Failure isn't fatal, but we still
961 * log an error message */
962 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
963 la->la_uid, 1, oh->ot_tx);
965 CERROR("%s: failed to update accounting ZAP for user "
966 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
967 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
968 obj->oo_attr.la_uid, -1, oh->ot_tx);
970 CERROR("%s: failed to update accounting ZAP for user "
971 "%d (%d)\n", osd->od_svname,
972 obj->oo_attr.la_uid, rc);
974 if ((la->la_valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
975 /* Update group accounting. Failure isn't fatal, but we still
976 * log an error message */
977 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
978 la->la_gid, 1, oh->ot_tx);
980 CERROR("%s: failed to update accounting ZAP for user "
981 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
982 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
983 obj->oo_attr.la_gid, -1, oh->ot_tx);
985 CERROR("%s: failed to update accounting ZAP for user "
986 "%d (%d)\n", osd->od_svname,
987 obj->oo_attr.la_gid, rc);
990 write_lock(&obj->oo_attr_lock);
992 if (la->la_valid & LA_ATIME) {
993 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
994 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL,
997 if (la->la_valid & LA_MTIME) {
998 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
999 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL,
1002 if (la->la_valid & LA_CTIME) {
1003 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1004 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL,
1007 if (la->la_valid & LA_MODE) {
1008 /* mode is stored along with type, so read it first */
1009 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1010 (la->la_mode & ~S_IFMT);
1011 osa->mode = obj->oo_attr.la_mode;
1012 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL,
1015 if (la->la_valid & LA_SIZE) {
1016 osa->size = obj->oo_attr.la_size = la->la_size;
1017 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL,
1020 if (la->la_valid & LA_NLINK) {
1021 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1022 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL,
1025 if (la->la_valid & LA_RDEV) {
1026 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1027 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL,
1030 if (la->la_valid & LA_FLAGS) {
1031 osa->flags = attrs_fs2zfs(la->la_flags);
1032 /* many flags are not supported by zfs, so ensure a good cached
1034 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1035 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL,
1038 if (la->la_valid & LA_UID) {
1039 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1040 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL,
1043 if (la->la_valid & LA_GID) {
1044 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1045 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL,
1048 obj->oo_attr.la_valid |= la->la_valid;
1049 write_unlock(&obj->oo_attr_lock);
1051 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1053 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1060 * XXX temporary solution.
1063 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1064 struct dt_object *parent, struct dt_object *child,
1065 cfs_umode_t child_mode)
1069 memset(ah, 0, sizeof(*ah));
1070 ah->dah_parent = parent;
1071 ah->dah_mode = child_mode;
1074 static int osd_declare_object_create(const struct lu_env *env,
1075 struct dt_object *dt,
1076 struct lu_attr *attr,
1077 struct dt_allocation_hint *hint,
1078 struct dt_object_format *dof,
1079 struct thandle *handle)
1081 char *buf = osd_oti_get(env)->oti_str;
1082 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1083 struct osd_object *obj = osd_dt_obj(dt);
1084 struct osd_device *osd = osd_obj2dev(obj);
1085 struct osd_thandle *oh;
1092 switch (dof->dof_type) {
1096 if (obj->oo_dt.do_body_ops == NULL)
1097 obj->oo_dt.do_body_ops = &osd_body_ops;
1103 LASSERT(handle != NULL);
1104 oh = container_of0(handle, struct osd_thandle, ot_super);
1105 LASSERT(oh->ot_tx != NULL);
1107 switch (dof->dof_type) {
1110 /* for zap create */
1111 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1116 /* first, we'll create new object */
1117 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1125 /* and we'll add it to some mapping */
1126 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1127 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1128 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1130 /* we will also update inode accounting ZAPs */
1131 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1132 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1133 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1134 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1136 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1138 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1139 XATTR_NAME_LMA, oh);
1141 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1142 false, NULL, false);
1146 int __osd_attr_init(const struct lu_env *env, udmu_objset_t *uos, uint64_t oid,
1147 dmu_tx_t *tx, struct lu_attr *la, uint64_t parent)
1149 sa_bulk_attr_t *bulk;
1150 sa_handle_t *sa_hdl;
1151 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1159 gen = dmu_tx_get_txg(tx);
1161 ZFS_TIME_ENCODE(&now, crtime);
1163 osa->atime[0] = la->la_atime;
1164 osa->ctime[0] = la->la_ctime;
1165 osa->mtime[0] = la->la_mtime;
1166 osa->mode = la->la_mode;
1167 osa->uid = la->la_uid;
1168 osa->gid = la->la_gid;
1169 osa->rdev = la->la_rdev;
1170 osa->nlink = la->la_nlink;
1171 osa->flags = attrs_fs2zfs(la->la_flags);
1172 osa->size = la->la_size;
1174 /* Now add in all of the "SA" attributes */
1175 rc = -sa_handle_get(uos->os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1179 OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1185 * we need to create all SA below upon object create.
1187 * XXX The attribute order matters since the accounting callback relies
1188 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1189 * look up the UID/GID attributes. Moreover, the callback does not seem
1190 * to support the spill block.
1191 * We define attributes in the same order as SA_*_OFFSET in order to
1192 * work around the problem. See ORI-610.
1195 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(uos), NULL, &osa->mode, 8);
1196 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(uos), NULL, &osa->size, 8);
1197 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(uos), NULL, &gen, 8);
1198 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(uos), NULL, &osa->uid, 8);
1199 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(uos), NULL, &osa->gid, 8);
1200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(uos), NULL, &parent, 8);
1201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(uos), NULL, &osa->flags, 8);
1202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(uos), NULL, osa->atime, 16);
1203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(uos), NULL, osa->mtime, 16);
1204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(uos), NULL, osa->ctime, 16);
1205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(uos), NULL, crtime, 16);
1206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(uos), NULL, &osa->nlink, 8);
1207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(uos), NULL, &osa->rdev, 8);
1209 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1211 OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1213 sa_handle_destroy(sa_hdl);
1218 * The transaction passed to this routine must have
1219 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1220 * to a transaction group.
1222 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
1223 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1224 uint64_t parent, void *tag)
1230 spin_lock(&uos->lock);
1232 spin_unlock(&uos->lock);
1234 /* Assert that the transaction has been assigned to a
1235 transaction group. */
1236 LASSERT(tx->tx_txg != 0);
1238 /* Create a new DMU object. */
1239 oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1240 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1241 rc = -sa_buf_hold(uos->os, oid, tag, dbp);
1245 LASSERT(la->la_valid & LA_MODE);
1249 return __osd_attr_init(env, uos, oid, tx, la, parent);
1253 * The transaction passed to this routine must have
1254 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1255 * to a transaction group.
1257 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1258 * This is fine for directories today, because storing the FID in the dirent
1259 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1260 * then we might need to re-evaluate the use of this flag and instead do
1261 * a conversion from the different internal ZAP hash formats being used. */
1262 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
1263 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1264 struct lu_attr *la, uint64_t parent,
1265 void *tag, zap_flags_t flags)
1272 spin_lock(&uos->lock);
1274 spin_unlock(&uos->lock);
1276 /* Assert that the transaction has been assigned to a
1277 transaction group. */
1278 LASSERT(tx->tx_txg != 0);
1280 oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
1281 DMU_OT_DIRECTORY_CONTENTS, 12, 12,
1282 DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1284 rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
1288 LASSERT(la->la_valid & LA_MODE);
1292 return __osd_attr_init(env, uos, oid, tx, la, parent);
1295 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1296 struct lu_attr *la, uint64_t parent,
1297 struct osd_thandle *oh)
1302 /* Index file should be created as regular file in order not to confuse
1303 * ZPL which could interpret them as directory.
1304 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1306 LASSERT(S_ISREG(la->la_mode));
1307 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1308 parent, osd_obj_tag, ZAP_FLAG_UINT64_KEY);
1314 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1315 struct lu_attr *la, uint64_t parent,
1316 struct osd_thandle *oh)
1321 LASSERT(S_ISDIR(la->la_mode));
1322 rc = __osd_zap_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1323 parent, osd_obj_tag, 0);
1329 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1330 struct lu_attr *la, uint64_t parent,
1331 struct osd_thandle *oh)
1336 LASSERT(S_ISREG(la->la_mode));
1337 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1338 parent, osd_obj_tag);
1343 * XXX: a hack, OST to use bigger blocksize. we need
1344 * a method in OSD API to control this from OFD/MDD
1346 if (!lu_device_is_md(osd2lu_dev(osd))) {
1347 rc = -dmu_object_set_blocksize(osd->od_objset.os,
1349 128 << 10, 0, oh->ot_tx);
1351 CERROR("%s: can't change blocksize: %d\n",
1352 osd->od_svname, rc);
1360 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1361 struct lu_attr *la, uint64_t parent,
1362 struct osd_thandle *oh)
1367 LASSERT(S_ISLNK(la->la_mode));
1368 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1369 parent, osd_obj_tag);
1375 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1376 struct lu_attr *la, uint64_t parent,
1377 struct osd_thandle *oh)
1382 la->la_valid = LA_MODE;
1383 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1384 la->la_valid |= LA_RDEV;
1386 rc = __osd_object_create(env, &osd->od_objset, &db, oh->ot_tx, la,
1387 parent, osd_obj_tag);
1393 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1394 struct osd_device *osd,
1397 struct osd_thandle *oh);
1399 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1401 osd_obj_type_f result;
1427 * Primitives for directory (i.e. ZAP) handling
1429 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1430 const struct lu_fid *fid, struct osd_thandle *oh)
1432 struct osd_thread_info *info = osd_oti_get(env);
1433 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1437 lustre_lma_init(lma, fid, 0);
1438 lustre_lma_swab(lma);
1440 buf.lb_len = sizeof(*lma);
1442 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1443 LU_XATTR_CREATE, oh, BYPASS_CAPA);
1449 * Concurrency: @dt is write locked.
1451 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1452 struct lu_attr *attr,
1453 struct dt_allocation_hint *hint,
1454 struct dt_object_format *dof,
1457 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1458 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1459 struct osd_object *obj = osd_dt_obj(dt);
1460 struct osd_device *osd = osd_obj2dev(obj);
1461 char *buf = osd_oti_get(env)->oti_str;
1462 struct osd_thandle *oh;
1469 /* concurrent create declarations should not see
1470 * the object inconsistent (db, attr, etc).
1471 * in regular cases acquisition should be cheap */
1472 down(&obj->oo_guard);
1474 LASSERT(osd_invariant(obj));
1475 LASSERT(!dt_object_exists(dt));
1476 LASSERT(dof != NULL);
1478 LASSERT(th != NULL);
1479 oh = container_of0(th, struct osd_thandle, ot_super);
1482 * XXX missing: Quote handling.
1485 LASSERT(obj->oo_db == NULL);
1487 /* to follow ZFS on-disk format we need
1488 * to initialize parent dnode properly */
1490 if (hint && hint->dah_parent)
1491 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1493 db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1495 GOTO(out, rc = PTR_ERR(th));
1498 zde->zde_dnode = db->db_object;
1499 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1501 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1503 rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
1507 /* Add new object to inode accounting.
1508 * Errors are not considered as fatal */
1509 rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
1510 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1513 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1514 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1515 rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
1516 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1519 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1520 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1522 /* configure new osd object */
1524 rc = osd_object_init0(env, obj);
1525 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1526 LASSERT(osd_invariant(obj));
1528 rc = osd_init_lma(env, obj, fid, oh);
1530 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1531 osd->od_svname, PFID(fid), rc);
1532 /* ignore errors during LMA initialization */
1541 static int osd_declare_object_ref_add(const struct lu_env *env,
1542 struct dt_object *dt,
1545 return osd_declare_attr_set(env, dt, NULL, th);
1549 * Concurrency: @dt is write locked.
1551 static int osd_object_ref_add(const struct lu_env *env,
1552 struct dt_object *dt,
1553 struct thandle *handle)
1555 struct osd_object *obj = osd_dt_obj(dt);
1556 struct osd_thandle *oh;
1557 struct osd_device *osd = osd_obj2dev(obj);
1558 udmu_objset_t *uos = &osd->od_objset;
1564 LASSERT(osd_invariant(obj));
1565 LASSERT(dt_object_exists(dt));
1566 LASSERT(obj->oo_sa_hdl != NULL);
1568 oh = container_of0(handle, struct osd_thandle, ot_super);
1570 write_lock(&obj->oo_attr_lock);
1571 nlink = ++obj->oo_attr.la_nlink;
1572 write_unlock(&obj->oo_attr_lock);
1574 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1578 static int osd_declare_object_ref_del(const struct lu_env *env,
1579 struct dt_object *dt,
1580 struct thandle *handle)
1582 return osd_declare_attr_set(env, dt, NULL, handle);
1586 * Concurrency: @dt is write locked.
1588 static int osd_object_ref_del(const struct lu_env *env,
1589 struct dt_object *dt,
1590 struct thandle *handle)
1592 struct osd_object *obj = osd_dt_obj(dt);
1593 struct osd_thandle *oh;
1594 struct osd_device *osd = osd_obj2dev(obj);
1595 udmu_objset_t *uos = &osd->od_objset;
1601 LASSERT(osd_invariant(obj));
1602 LASSERT(dt_object_exists(dt));
1603 LASSERT(obj->oo_sa_hdl != NULL);
1605 oh = container_of0(handle, struct osd_thandle, ot_super);
1606 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1608 write_lock(&obj->oo_attr_lock);
1609 nlink = --obj->oo_attr.la_nlink;
1610 write_unlock(&obj->oo_attr_lock);
1612 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(uos), &nlink, 8, oh);
1616 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1617 struct lustre_capa *capa, struct lustre_capa_key *keys)
1619 struct osd_thread_info *oti = osd_oti_get(env);
1620 struct obd_capa *oc;
1624 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1626 if (capa_is_expired(oc)) {
1627 DEBUG_CAPA(D_ERROR, capa, "expired");
1634 spin_lock(&capa_lock);
1635 for (i = 0; i < 2; i++) {
1636 if (keys[i].lk_keyid == capa->lc_keyid) {
1637 oti->oti_capa_key = keys[i];
1641 spin_unlock(&capa_lock);
1644 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1648 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1651 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1653 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1657 oc = capa_add(dev->od_capa_hash, capa);
1663 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1664 struct lustre_capa *capa, __u64 opc)
1666 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1667 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1670 if (!dev->od_fl_capa)
1673 if (capa == BYPASS_CAPA)
1677 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1681 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1682 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1686 if (!capa_opc_supported(capa, opc)) {
1687 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1691 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1692 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1699 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1700 struct dt_object *dt,
1701 struct lustre_capa *old,
1704 struct osd_thread_info *info = osd_oti_get(env);
1705 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1706 struct osd_object *obj = osd_dt_obj(dt);
1707 struct osd_device *dev = osd_obj2dev(obj);
1708 struct lustre_capa_key *key = &info->oti_capa_key;
1709 struct lustre_capa *capa = &info->oti_capa;
1710 struct obd_capa *oc;
1714 if (!dev->od_fl_capa)
1715 RETURN(ERR_PTR(-ENOENT));
1717 LASSERT(dt_object_exists(dt));
1718 LASSERT(osd_invariant(obj));
1720 /* renewal sanity check */
1721 if (old && osd_object_auth(env, dt, old, opc))
1722 RETURN(ERR_PTR(-EACCES));
1724 capa->lc_fid = *fid;
1727 capa->lc_flags = dev->od_capa_alg << 24;
1728 capa->lc_timeout = dev->od_capa_timeout;
1729 capa->lc_expiry = 0;
1731 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1733 LASSERT(!capa_is_expired(oc));
1737 spin_lock(&capa_lock);
1738 *key = dev->od_capa_keys[1];
1739 spin_unlock(&capa_lock);
1741 capa->lc_keyid = key->lk_keyid;
1742 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1744 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1746 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1748 RETURN(ERR_PTR(rc));
1751 oc = capa_add(dev->od_capa_hash, capa);
1755 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
1757 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1760 /* XXX: no other option than syncing the whole filesystem until we
1762 txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
1767 static struct dt_object_operations osd_obj_ops = {
1768 .do_read_lock = osd_object_read_lock,
1769 .do_write_lock = osd_object_write_lock,
1770 .do_read_unlock = osd_object_read_unlock,
1771 .do_write_unlock = osd_object_write_unlock,
1772 .do_write_locked = osd_object_write_locked,
1773 .do_attr_get = osd_attr_get,
1774 .do_declare_attr_set = osd_declare_attr_set,
1775 .do_attr_set = osd_attr_set,
1776 .do_ah_init = osd_ah_init,
1777 .do_declare_create = osd_declare_object_create,
1778 .do_create = osd_object_create,
1779 .do_declare_destroy = osd_declare_object_destroy,
1780 .do_destroy = osd_object_destroy,
1781 .do_index_try = osd_index_try,
1782 .do_declare_ref_add = osd_declare_object_ref_add,
1783 .do_ref_add = osd_object_ref_add,
1784 .do_declare_ref_del = osd_declare_object_ref_del,
1785 .do_ref_del = osd_object_ref_del,
1786 .do_xattr_get = osd_xattr_get,
1787 .do_declare_xattr_set = osd_declare_xattr_set,
1788 .do_xattr_set = osd_xattr_set,
1789 .do_declare_xattr_del = osd_declare_xattr_del,
1790 .do_xattr_del = osd_xattr_del,
1791 .do_xattr_list = osd_xattr_list,
1792 .do_capa_get = osd_capa_get,
1793 .do_object_sync = osd_object_sync,
1796 static struct lu_object_operations osd_lu_obj_ops = {
1797 .loo_object_init = osd_object_init,
1798 .loo_object_delete = osd_object_delete,
1799 .loo_object_release = osd_object_release,
1800 .loo_object_free = osd_object_free,
1801 .loo_object_print = osd_object_print,
1802 .loo_object_invariant = osd_object_invariant,
1805 static int osd_otable_it_attr_get(const struct lu_env *env,
1806 struct dt_object *dt,
1807 struct lu_attr *attr,
1808 struct lustre_capa *capa)
1814 static struct dt_object_operations osd_obj_otable_it_ops = {
1815 .do_attr_get = osd_otable_it_attr_get,
1816 .do_index_try = osd_index_try,